A real-time AI coaching application that analyzes your fighting form using computer vision and machine learning. This application acts as a virtual coach, providing instant feedback on your technique for various strikes.

• Real-time Pose Detection: Utilizes MediaPipe to track 33 key body landmarks in real-time.
• Move Classification: Automatically identifies moves including:
  • Jab
  • Cross
  • Uppercut
  • Kick
  • Knee
  • Neutral Stance
• Instant Feedback: Provides corrective feedback on your form (e.g., "Extend arm more", "Keep elbow bent", "Rotate hips") based on biomechanical analysis.
• Voice Control: Hands-free operation with voice commands:
  • "Start Coach" to begin a session.
  • "End Coach" to stop and review.
• Session Recording: Automatically records your session and provides a replay with synchronized feedback logs.
• Visual Overlay: Displays feedback and status directly on the video feed.

1. Clone the repository:

   git clone <repository-url>
   cd <repository-directory>

2. Install dependencies:

   pip install -r requirements.txt

   Note: Ensure you have flask, pandas, scikit-learn, joblib, opencv-python, and mediapipe installed.

1. Start the application:

   python app.py

2. Open the interface: Open your web browser and navigate to http://localhost:5001.

3. Start Training:

   • Allow camera and microphone permissions when prompted.
   • Stand back so your full body is visible in the camera frame.
   • Click the Start Coach button or say "Start Coach".

4. Receive Feedback:

   • Perform strikes (jabs, crosses, etc.).
   • Read the real-time feedback on the screen.
   • Listen for audio cues (if enabled).

5. Review Session:

   • Click End Coach or say "End Coach".
   • A modal will appear with a video replay of your session and a detailed log of all feedback received.

• Backend: Python, Flask
• Computer Vision: MediaPipe, OpenCV
• Machine Learning: Scikit-learn (Random Forest Classifier), Pandas, NumPy
• Frontend: HTML5, CSS3, JavaScript (Canvas API, MediaRecorder API, Web Speech API)

The main entry point for the Flask application. It handles:

• Routing: Defines endpoints for the web interface (/) and API routes for session management (/start_session, /end_session) and pose processing (/process_pose).
• Session Management: Manages active user sessions and stores temporary pose data using a sliding window approach.
• Data Flow: Receives landmarks from the frontend, forwards them to the backend logic, and returns real-time feedback.

Contains the core logic for the application, including:

• Feature Extraction: Converts raw MediaPipe landmarks into biomechanical features (angles, velocities, relative positions).
• Motion Analysis: Implements the build_features_from_window function to create a feature vector from a sequence of frames.
• Feedback Generation: The generate_feedback function analyzes the predicted move and specific feature values to produce actionable, biomechanically accurate advice (e.g., correcting arm extension or guard position).
• Model Loading: Loads the pre-trained Random Forest classifier and MediaPipe configuration.

The frontend user interface, built with HTML5, CSS3, and JavaScript. Key components include:

• Video Processing: Uses the MediaPipe Pose JS solution to detect landmarks directly in the browser.
• Canvas Rendering: Draws the video feed, landmarks, and overlay feedback on an HTML5 Canvas.
• Session Control: Manages the recording state, handles the "Start/End Coach" buttons, and captures video for replay using the MediaRecorder API.
• Voice Recognition: Integrates the Web Speech API for hands-free voice commands.

A directory containing the machine learning assets:

• Random Forest Model: The trained sklearn model used to classify martial arts moves.
• Pose Landmarker: The MediaPipe model file used for pose detection (if running in backend mode, though the frontend also uses the JS version).